Convolutional neural network architecture for beam instabilities identification in Synchrotron Radiation Systems as an anomaly detection problem

• Published in: Measurement : journal of the International Measurement Confederation 2020-12, Vol.165, p.108116, Article 108116
• Main Authors: Piekarski, Michał, Jaworek-Korjakowska, Joanna, Wawrzyniak, Adriana I., Gorgon, Marek
• Format: Article
• Language: English
• Subjects: Anomalies; Anomaly detection; Artificial neural networks; Complex systems; Computer architecture; Convolutional codes; Convolutional neural network; Fault detection; Fault diagnosis; Neural networks; Sensors; Signal analysis; Subsystems; Synchrotron radiation; Synchrotron radiation system; Synchrotrons; Transfer learning; Windows (intervals)
• ISSN: 0263-2241; 1873-412X
• DOI: 10.1016/j.measurement.2020.108116

•We present an automatic fault detection system using pretrained CNN network.•First study on CNN techniques for anomaly detection in Synchrotron Radiation Centre.•CNN architectures with adjusted classifier are compared for beam stability analysis.•Study has been performed on 336 h of 64 SIP signals located around the storage ring.•VGG-16 architecture performed best and achieved 92% accuracy. Solaris National Synchrotron Radiation Centre is a research facility that provides high quality synchrotron light. To control such a complex system it is necessary to monitor signals from various devices and subsystems. Despite the high demand for solutions to monitor the operation of centres, little work has concentrated on automatic analysis and fault detection. Anomaly detection prevents from financial loss, unplanned downtimes and in extreme cases cause damage. To address the problem a convolutional neural network (CNN) for fault detection in time series data has been proposed. The aim of the system is to identify abnormal status of sensors in certain time steps. In this study, we deploy transfer learning by examining pre-trained VGG-16, VGG-19, InceptionV3 and Xception CNN models with an adjusted densely-connected classifiers. Our database contains 336 h of signals in total which have been divided into 6300 time windows of 3 min length. The proposed solution, based on the VGG-16 architecture, detects anomalies in diagnostics signals with 92% accuracy and 85.5% precision what is a state-of-the-art result.